Mailman 3 June 2002 - Cplusplus-sig

RE: Cygwin Boost.Python Build
by Raoul Gough Aug. 3, 2002

Aug. 3, 2002

If it's any help, here's my set-up script for building Boost with the Cygwin Python: export PYTHON='f:\cygwin\bin\python.exe' export PYTHON_ROOT='f:\cygwin\lib\python2.1' export PYTHON_VERSION='2.1' export PYTHON_INCLUDES='f:\cygwin\usr\include\python2.1' export PYTHON_LIB_PATH='f:\cygwin\lib\python2.1\config' export PYTHON_STDLIB_PATH='f:\cygwin\lib\python2.1' export GCC_PYTHON_ROOT='f:\cygwin\usr' I remember it taking some time to arrive at this - maybe it will help Scott get his set-up working (with appropriate replacements). Regards, Raoul Gough. _________________________________________________________ Do You Yahoo!? Get your free @yahoo.com address at http://mail.yahoo.com

2 2

Boost.Python v2: beginning of generalized object support
by David Abrahams July 19, 2002

July 19, 2002

I've just checked in the foundations for generalized object support (boost/python/object.hpp). Details: borrow() has been renamed to borrowed() as per user suggestions. handle<> now converts automatically to/from python in argument lists and return values object converts automatically to/from python also object attribute access is available through the "_" member function: C++ Python x._("foo") = 1; x.foo = 1 y = x._("foo"); y = x.foo getitem/setitem support is implemented via operator[] in the expected ways. x["foo"] = 1; x['foo'] = 1 y = x[z]; y = x[z]; callable objects can be called from C++, just as from Python: z = x(1, y) z = x(1, y) operators are coming, as are the python built-in functions such as isinstance(). Regards, Dave +---------------------------------------------------------------+ David Abrahams C++ Booster (http://www.boost.org) O__ == Pythonista (http://www.python.org) c/ /'_ == resume: http://users.rcn.com/abrahams/resume.html (*) \(*) == email: david.abrahams(a)rcn.com +---------------------------------------------------------------+

5 34

Boost.Python V2: list implementation
by David Abrahams July 3, 2002

July 3, 2002

I have just checked in the implementation of the Python list class wrapper! Tests in libs/python/test/list.[cpp/py]. This opens up the possibility of implementing the other built-in types: tuple, dict, str, int_, long_, et. al. This is another of those "fairly mechanical" tasks with which I would really appreciate having some help. The interface design follows the following protocol: I did: >>> help(list) >From within Python and looked at the output, ignoring all the "special" (__<name>__) methods, which are handled already by class object. Where arguments can be any Python type, I provided a template member function to cover that (e.g. list.index()) and a version in the dll/so which accepts object. Where arguments must be a specific type (e.g. pop(x)), I provided an overload taking the specific type and another accepting object, which will allow it to be used in expressions involving object. As we get these other types implemented we can carve them out of objects2.[cpp/hpp]. I hope someone will take up this challenge, and that this description will be helpful. Enjoy, Dave +---------------------------------------------------------------+ David Abrahams C++ Booster (http://www.boost.org) O__ == Pythonista (http://www.python.org) c/ /'_ == resume: http://users.rcn.com/abrahams/resume.html (*) \(*) == email: david.abrahams(a)rcn.com +---------------------------------------------------------------+

2 2

high level overview of boost.python implementation
by Achim Domma July 1, 2002

July 1, 2002

Hi Dave, would it be possible to get a high level overview ot the implementation of boost.python? (if there's already such a document, I was not able to find it) You sometimes ask for contributions, especially implementation of not too complex, straight forward code. Looking at the code, I understand single pieces, but I do not get the big picture, which is neccessary, if I would try to contribute (and I would like to contribute, if possible). I don't ask for more documentation, only a starting point where look, to get familiar with the code, and the main concepts you are using. Achim

4 9

python sequence to stl container
by Martin Casado June 28, 2002

June 28, 2002

> ----- Original Message ----- > From: "Martin Casado" <casado2(a)llnl.gov> > > > > Ralf's version is the one to use, for now. I'm still hoping to get him > > to > > > > build the one we really want to put in the library... > > > > If there is a description for the one you really want to put into the > > library I can give it a stab.. > > We've been discussing a version which converts any sequence without going > through an intermediate tuple. Part of this requires changing the rvalue > conversion protocol so that the convertible() function is able to do some > or all of the conversion work, by giving it access to the > rvalue_from_python_data. In the case of Python sequences, you really have > to try to do all the conversion as part of the overload resolution step > since you might not be able to make a second pass over some sequences (e.g. > input streams). Is it not possible to change the rvalue conversion protocol so that the check to see if the object is convertible and the conversion happen within construct() since it already has access to the rvalue_from_python_data? The problem I suppose is signifying that the object is not convertible, using exceptions would almost certainly be too slow. Do you have some ideas about how you were going to go about this? ~~m

2 1

return type of list::index(object const&)
by David Abrahams June 28, 2002

June 28, 2002

I have a choice about whether to return python::object or long from this function and others like it. I'm leaning towards long; anyone want to argue for object? -Dave +---------------------------------------------------------------+ David Abrahams C++ Booster (http://www.boost.org) O__ == Pythonista (http://www.python.org) c/ /'_ == resume: http://users.rcn.com/abrahams/resume.html (*) \(*) == email: david.abrahams(a)rcn.com +---------------------------------------------------------------+

2 2

Another BPLV2 newbie...
by Hugo van der Merwe June 27, 2002

June 27, 2002

I recently finally got round to taking a look at V2, and seeing if I can do something with it. (Seems I can't ;) I immediately ran into two things I don't know how to handle. First is related to trying to wrap a library which has constructors with many many parameters. With V1 the solution was: python ../gen_all.py 20 20 isn't quite necessary. How do I do something similar with V2? Second, what is wrong with this: ==== test.cpp ==== #define BOOST_PYTHON_V2 #include <boost/python/module.hpp> class test { public: test(); }; BOOST_PYTHON_MODULE_INIT(testmod) { boost::python::module testmod("testmod"); testmod .add( boost::python::class_<test>("test") .def_init() ); } ==== test.cpp ==== I get the following error tryinc to compile it: demeter.cpp:42: type `boost::python::class_<Demeter::Settings,boost::python::detail::not_specified,boost::python::detail::not_specified,boost::python::detail::not_specified>' is not yet defined I'm worried I might have broken something by trying gen_all.py, but the tests appear to build fine. (I ran $ jam -sPYTHON_VERSION=2.2 -sPYTHON_ROOT=/usr in libs/python/test, with BOOST_ROOT set in environment.) Thanks, Hugo van der Merwe

6 7

tuple, list, dict, et. al
by David Abrahams June 26, 2002

June 26, 2002

The answers below are mine, but I'm interested in feedback about them. ---- Question: What should the following do, where o is a python::object? tuple(o) Answer: I believe (strongly) that it should do something exactly corresponding to what does in Python: >>> help(tuple) Help on class tuple in module __builtin__: class tuple(object) | tuple() -> an empty tuple | tuple(sequence) -> tuple initialized from sequence's items | | If the argument is a tuple, the return value is the same object. Question: How do I specify that a wrapped function accepts a tuple argument? Answer: Put a boost::tuple in the function signature Question: What should that mean in terms of overload resolution? Answer 1: The function only matches if the actual argument is a Python tuple or a subclass of tuple Answer 2: The function matches if the actual argument is any Python sequence; tuple(arg) will be applied to retrieve the tuple object used inside the wrapped function Which of the above answers is better? So far liberal conversion during function matching (e.g. Python Float -> C++ int) has caused problems for at least one person. On the other hand, in the long run that problem should be resolved and overloads should be resolved more-smartly. Opinions? -Dave +---------------------------------------------------------------+ David Abrahams C++ Booster (http://www.boost.org) O__ == Pythonista (http://www.python.org) c/ /'_ == resume: http://users.rcn.com/abrahams/resume.html (*) \(*) == email: david.abrahams(a)rcn.com +---------------------------------------------------------------+

4 11

RE: [C++-sig] conversion of std::list<...>
by Ralf W. Grosse-Kunstleve June 26, 2002

June 26, 2002

> Cool! Please remind me of this on Wednesday; I am currently stuck in a > Microsoft gravity well, trying to survive the inevitable impact. The struct register_container_from_python_sequence in the bpl_utils still has an undesirable feature: the input Python sequence is traversed twice. This does not matter if the user passes a list or tuple, but will cause confusion (at least) if the user passes an object that can only be traversed once (something similar to a file object for example; or some kind of iterator). Consider the static member functions of register_container_from_python_sequence: static void* convertible(PyObject* obj_ptr) { //... sequence_fast seq(obj, 0); // first traversal //... } static void construct(PyObject* obj_ptr, boost::python::converter::rvalue_stage1_data* data) { //... sequence_fast seq(obj, 0); // second traversal //... } Could we somehow avoid the second traversal? Idea: sequence_fast has a handle<> with the result of PySequence_Fast(). I know I could return the corresponding PyObject* from convertible(), and this will then be passed as the first argument to construct(). I.e. I could just use the PyObject* knowing it is the result of PySequence_Fast() and would not have to call PySequence_Fast() a second time. Problem: PySequence_Fast returns a new reference. There is no guarantee that construct() will be called. So the reference count has to be managed outside of convertible() and construct(). Could this be done in the current framework? Thanks, Ralf __________________________________________________ Do You Yahoo!? Yahoo! - Official partner of 2002 FIFA World Cup http://fifaworldcup.yahoo.com

2 1

Profiling object instantiation... a preliminary report
by Greg Landrum June 26, 2002

June 26, 2002

Re-establishing context: I have an extension class that I've wrapped using both Boost and Cxx. I need to be able to pickle and depickle large numbers (1e5-1e6) of these objects quickly for my target application (storing them in a ZODB and querying the resulting database). I discovered that simple instantiation of 1e5 objects took dramatically longer in Boost than in Cxx (6.7 seconds vs 0.9 seconds). I pulled down a copy of Intel's vtune profiler[1] and profiled my Boost and Cxx wrapped libraries to find the origin of the huge performance difference. After fiddling around for a while figuring out the appropriate magic to make vtune actually work, I found something surprising. I had exposed 4 different constructors for my Mol class. What I failed to realize is that when you overload functions (at least constructors) in BPL, this is handled using exceptions. This was at least partially responsible for the tremendous slow down in the Boost wrapper of my code. When I commented out some constructors, rebuilt, and reprofiled my code I got a run time of 1.9 seconds, only twice the Cxx run time. I'm a lot happier with this, but I'll continue to do a bit of exploration to see if I can bring that down more. I'm rather enjoying vtune now that I've figure out what can be done with it. If anyone wants more detailed information, I'd be happy to provide it. -greg [1] The beta of vtune is available as a free download for linux and windows. These results are under Win2K using MSVC++ v6. ---- greg Landrum (greglandrum(a)earthlink.net) Software Carpenter/Computational Chemist

6 12